Image-Based Evaluation of Fiber volume fraction in Composite Cross Sections

Abstract

This thesis introduces an innovative, non-destructive way for measuring fiber volume fraction in filament-wound composite tubes using microscopic image analysis and Python-based automation. Recognizing fiber volume fraction's crucial role in defining composite mechanical behavior, the study proposes a digital technique that outperforms traditional destructive testing in terms of precision, repeatability, and efficiency. A carefully structured procedure was carried out on glass fiber composite samples provided by L-Tec Sport, including sample polishing, multi-magnification optical microscopy, and greyscale threshold-based segmentation. Python modules such as OpenCV, NumPy, and Matplotlib were used to classify fibers, matrix, and voids, allowing for pixel-level examination of microstructure with minimal input from users. What highlights this work is the combination of statistical fiber diameter analysis (from over 200 hand-measured fibers, ranged from 29 to 39 pixels) with automated pixel-based fiber volume fraction computation, resulting in findings that nearly match manufacturer data 50-60%. The Python-derived fiber volume fraction values varied from 52% to 55%, while hand computations produced 46-54%. A ±1 pixel inaccuracy in fiber diameter may affect area estimations by up to 77.8% at lower magnifications (magnification of 25X), emphasizing the importance of balancing resolution and accuracy at 100X magnification, the error percentage reduces to less than 5%, indicating the lowest measured error. Edge-based approaches, such as Canny detection, were investigated, but they failed to provide a reliable result. This research is mix of digital image analysis and engineering knowledge establishes a compelling new benchmark for composite material evaluation.